# Data Lineage for a Small Business Report: Trace One Field From Source to Decision

<a id="main-content"></a>

Operations &amp; Systems

October 28, 2026  ·  8 min read  ·  Victor Valentine Romo

On this page

1. [Separate the Artifact, the Activity, and the Responsible Agent](<#separate-the-artifact-the-activity-and-the-responsible-agent>)
2. [Give Every Input and Result a Version](<#give-every-input-and-result-a-version>)
3. [Write the Selection Before Calculating Duration](<#write-the-selection-before-calculating-duration>)
4. [Describe Conversions, Joins, and Derived Fields](<#describe-conversions-joins-and-derived-fields>)
5. [Trace the Aggregate Into the Chart](<#trace-the-aggregate-into-the-chart>)
6. [Copy the Lineage Worksheet](<#copy-the-lineage-worksheet>)
7. [Keep Broken Links and Unknown Steps Visible](<#keep-broken-links-and-unknown-steps-visible>)
8. [Revise the Trace When the Rule Changes](<#revise-the-trace-when-the-rule-changes>)
9. [Sources](<#sources>)

The chart says request duration increased. Nobody can identify which export supplied the closing time.

The [workbook](</articles/excel-date-systems-business-handoffs/>) opens. The formula calculates. The chart has a label. One missing connection still leaves the decision without an explainable input.

Trace one field from the displayed result back to its source. Record what happened between each pair of artifacts.

Data lineage mapping gives that trace a working structure. The method below adapts concepts from the [W3C PROV Model Primer](<https://www.w3.org/TR/prov-primer/>) into a small reporting worksheet. It is an editorial proposal, not a W3C template or a claim of PROV conformance.

<a id="separate-the-artifact-the-activity-and-the-responsible-agent"></a>

## Separate the Artifact, the Activity, and the Responsible Agent

The [PROV primer's overview](<https://www.w3.org/TR/prov-primer/>) distinguishes entities, activities, and agents. An entity is a thing, such as a file or chart. An activity produces or changes entities. An agent carries responsibility for an activity. Software can be an agent, alongside people and organizations.

For this worksheet, [give each artifact a version identifier](</articles/spreadsheet-version-control-business-decisions/>). Describe each transformation as an activity. Assign an accountable role to explain its execution.

The export is an artifact. Converting [timestamps](</articles/event-time-processing-time-business-reports/>) is an activity. The reporting owner is the person responsible for explaining that conversion.

Keep those columns separate. A file called `final-report.xlsx` identifies neither the rule that produced it nor the person who accepted the result.

Start with the question behind the chart. In this fictional example, an operations coordinator wants to review elapsed duration for four completed requests. The result supports a discussion about those records. It does not establish a staffing requirement or explain why durations differ.

Define duration before tracing it. A [business metric dictionary](</articles/business-metric-dictionary-template/>) records what a measure means. The lineage record identifies how a specific output acquired its value.

<a id="give-every-input-and-result-a-version"></a>

## Give Every Input and Result a Version

All records and artifacts below are invented. They describe one reporting exercise, not a customer system or an observed business result.

Use these artifact identifiers for the exercise:

| Artifact identifier | Contents | Version boundary |
| --- | --- | --- |
| E1 | `requests-export-v1.csv` | Six synthetic requests, captured after the reporting cutoff |
| E2 | `request-types-v1.csv` | One label for each of two request type codes |
| E3 | `request-durations-v1.csv` | Four selected requests and derived minutes |
| E4 | `duration-summary-v1.csv` | Counts, total minutes, and mean minutes by type |
| E5 | `duration-chart-v1.html` | Two bars built from E4 mean minutes |

Record capture time, reporting cutoff, and file version separately. [An export captured after a cutoff](</articles/late-arriving-data-report-revisions/>) can contain events outside the selected period.

Preserve the exact input used for the report under [its applicable access controls](</articles/small-team-access-review-evidence/>). A link to an export screen that generates today's records does not identify yesterday's input.

In PROV, a revision can be represented as a new entity related to an earlier entity. The [primer explains derivation and revision](<https://www.w3.org/TR/prov-primer/>). Apply that distinction here by giving revised artifacts new identifiers. Do not overwrite E1 while retaining a trace that describes its earlier contents.

The identifiers are local to this example. A business can use its existing document versions or run identifiers. The requirement is enough specificity to distinguish what was used.

<a id="write-the-selection-before-calculating-duration"></a>

## Write the Selection Before Calculating Duration

E1 contains these rows. [Every timestamp has a UTC offset](</articles/automation-time-zones-daylight-saving/>). The reporting interval starts at `2026-10-01T00:00:00Z` and ends before `2026-10-02T00:00:00Z`.

| Request | Type | Status | Opened, UTC | Closed, UTC |
| --- | --- | --- | --- | --- |
| R1 | A | closed | 2026-10-01 09:00Z | 2026-10-01 09:30Z |
| R2 | A | closed | 2026-10-01 10:00Z | 2026-10-01 11:00Z |
| R3 | B | closed | 2026-10-01 12:00Z | 2026-10-01 13:30Z |
| R4 | B | closed | 2026-10-01 14:00Z | 2026-10-01 16:00Z |
| R5 | A | open | 2026-10-01 15:00Z | Missing |
| R6 | B | closed | 2026-10-01 23:00Z | 2026-10-02 00:30Z |

Rule P1 selects closed requests whose closing timestamp falls within that interval. R1 through R4 qualify. R5 has no closing event. R6 closes outside the interval.

Write the exclusion evidence beside the rule: six input rows, four selected rows, one open request, and one request outside the closing window.

Do not describe the filter as “October first requests.” That phrase leaves open whether the report selects opening dates or closing dates.

The example assumes E1 supplies exactly these six records. In a real report, trace the export's selection settings too. A downstream filter cannot recover [records the export omitted](</articles/api-pagination-business-export-completeness/>).

Keep [negative or missing durations](</articles/null-blank-zero-business-data/>) visible as exceptions under a stated rule. This example contains no negative duration among selected rows. That observation describes these records only.

<a id="describe-conversions-joins-and-derived-fields"></a>

## Describe Conversions, Joins, and Derived Fields

Rule P1 subtracts `opened_at` from `closed_at`, then divides elapsed seconds by sixty. The result field is `duration_minutes`.

The selected durations are thirty, sixty, ninety, and one hundred twenty minutes. Their total is three hundred minutes.

No business calendar applies. Nights, breaks, and working hours receive no special treatment. A report of working-time duration would need a different rule and another version.

E2 maps code A to “Routine” and code B to “Review.” P1 adds that label to each selected row. The example specifies exactly one matching label per type code.

Record the join key, join type, expected match count, and unmatched-row treatment. Here, a left join on `type_code` retains all four selected requests. Every selected request matches one label. No row multiplies.

Use the [join cardinality review](</articles/join-cardinality-business-reports/>) when a relationship can create multiple matches. Writing “joined the lookup” leaves the expected number of rows unexplained.

E3 contains four rows after the join. Preserve both the original type code and the added label. The code identifies the join input. The label identifies what the reader sees.

If the lookup later renames “Review,” keep E2 version one attached to this run. A changed label should not silently rewrite the explanation for an earlier chart.

<a id="trace-the-aggregate-into-the-chart"></a>

## Trace the Aggregate Into the Chart

Rule P2 groups E3 by type code and label. It counts requests, sums duration minutes, and divides each sum by its count.

| Type | Label | Requests | Total minutes | Mean minutes |
| --- | --- | --- | --- | --- |
| A | Routine | 2 | 90 | 45 |
| B | Review | 2 | 210 | 105 |

The arithmetic is inspectable: thirty plus sixty equals ninety. Ninety divided by two equals forty-five. Ninety plus one hundred twenty equals two hundred ten. Two hundred ten divided by two equals one hundred five.

Across all four requests, three hundred divided by four equals seventy-five minutes. That overall figure is a separate output. E5 displays only the two group means.

Rule P3 reads E4's `mean_minutes` field. It places the label on [the horizontal axis](</articles/business-chart-axis-scale-review/>) and minutes on the vertical axis. The chart title names the closing interval and the measure.

Record the displayed field even when the chart appears obvious. Selecting `total_minutes` would produce bars of ninety and two hundred ten. The source file could remain unchanged while the displayed meaning changes.

A chart screenshot records appearance. Preserve the chart configuration or equivalent field mapping as evidence for the activity. Keep an [audit of spreadsheet formulas](</articles/spreadsheet-formula-audit-business/>) separate when the aggregate comes from a workbook formula.

<a id="copy-the-lineage-worksheet"></a>

## Copy the Lineage Worksheet

Use one row for each derivation step. Split an activity when it has a different owner, rule, or result artifact.

This worksheet is the proposed adaptation. Its role names and identifiers belong to the fictional exercise.

| Output | Source artifact and field | Activity | Rule version | Result artifact | Owner | Evidence | Unresolved step |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Selected durations | E1: status, opened\_at, closed\_at, type\_code; E2: type\_code, label | Filter, subtract, convert, label | P1 v1 | E3 v1 | Reporting owner | Selection record; four retained rows; join match counts | None within this synthetic step |
| Group means | E3: type\_code, label, duration\_minutes | Group, count, sum, divide | P2 v1 | E4 v1 | Report reviewer | Two aggregate rows; arithmetic reconciliation | None within this synthetic step |
| Displayed bars | E4: label, mean\_minutes | Select fields and render | P3 v1 | E5 v1 | Chart owner | Field mapping; chart title and units | None within this synthetic step |
| Exported timestamps | Underlying collection record unknown | Capture and export | Unknown | E1 v1 | Source owner unassigned | Synthetic export only | Original collection and export implementation absent |

For a blank copy, retain these eight columns. Replace the example values with artifact identifiers, field names, named roles, and evidence locations.

Write the final output beside its use. Then work backward until the trace reaches a defined source boundary. Mark that boundary explicitly.

In this exercise, the worksheet reaches E1 but cannot establish how its timestamps entered the source system. That missing step remains visible.

An owner column identifies responsibility for answering a question. It does not prove the owner reviewed the evidence. Add the review date and outcome when that review occurs.

<a id="keep-broken-links-and-unknown-steps-visible"></a>

## Keep Broken Links and Unknown Steps Visible

Suppose someone changes R2's closing timestamp in a workbook before calculating E3. The current worksheet no longer explains the result.

Add the edited workbook as another artifact. Record the edit, its stated reason, the responsible role, and the supporting evidence. If the edit history is unavailable, mark the transformation unknown.

Do not reconstruct a convenient explanation from the final number. Several transformations can produce the same result. A matching total does not identify which one occurred.

Treat an inaccessible evidence link as inaccessible. Record its owner and the missing proof. An unavailable artifact and a confirmed incorrect transformation are different findings.

Lineage also leaves separate questions open. A perfectly described path can carry an incorrect source timestamp into every later result. [The path does not establish accuracy](</articles/data-quality-dimensions-business-reports/>).

Likewise, knowing where an export originated does not authorize access, copying, or publication. Keep evidence references within the permissions that apply to the underlying material.

Use a minimal evidence location when the worksheet circulates more widely than the data. Describe the controlled record and its custodian without reproducing restricted contents.

<a id="revise-the-trace-when-the-rule-changes"></a>

## Revise the Trace When the Rule Changes

Suppose the coordinator now wants requests opened within the interval. All six requests meet that date selection. R6 enters despite closing after the interval. R5 remains open, so [its completed duration is unknown](</articles/work-item-aging-small-service-teams/>).

Decide whether the revised report covers all opened requests or only completed requests opened during the interval. Record any completion filter and its exclusions. Keep R5's missing duration visible if it remains in the population. Do not substitute zero or silently reduce the [denominator](</articles/two-way-tables-business-reporting/>).

Create P1 version two and a new result artifact. Record the changed selection rule before comparing the revised chart with E5.

Keep the old report attached to its original trace. A later reader needs to know why two charts use different populations.

For the next run, check the source version, selection rule, conversion, join, aggregate, and chart field. Confirm each connection against its evidence.

Stop the trace where the evidence stops. Give the missing connection an owner. Keep the derivation explainable.

<a id="sources"></a>

## Sources

- [W3C: PROV Model Primer](<https://www.w3.org/TR/prov-primer/>). Working Group Note dated 2013.04.30. Concepts of entity, activity, agent, derivation, and revision. Checked 2026.10.28. The worksheet and reporting rules are proposed adaptations.
